BRIEF-DETAILS: Japanese CLIP model trained on 1B image-text pairs, featuring Eva02-B image encoder and BERT text encoder. Excels at image classification and retrieval tasks.
BRIEF DETAILS: FinTwitBERT-sentiment is a specialized 110M parameter BERT model fine-tuned for financial tweet sentiment analysis, trained on 1.4M+ tweets.
Brief-details: A distilled version of Whisper large-v3 with 809M parameters, optimized for faster inference while maintaining quality. Supports 99 languages for speech recognition.
Brief-details: Compact 1.1B parameter LLM fine-tuned on WizardVicuna dataset, offering balanced performance across multiple tasks with 34.76% average accuracy.
Brief-details: A 410M parameter language model from EleutherAI's Pythia suite, trained on The Pile dataset for research and interpretability studies. Features 24 layers and 1024 model dimension.
Brief Details: A specialized latent diffusion upscaler that doubles image resolution while maintaining quality, designed specifically for Stable Diffusion outputs.
BRIEF DETAILS: A powerful time series forecasting model with 205M parameters based on T5 architecture, offering 250x faster inference and improved accuracy over original Chronos models.
Brief Details: German-English bilingual text embedding model with 161M parameters, supporting 8192 token sequences and optimized for cross-lingual applications using ALiBi architecture
BRIEF-DETAILS: Vintage-style image generation model specializing in 1990s photography aesthetics with Kodachrome-like effects, light leaks, and nostalgic qualities
Brief-details: A powerful 6.7B parameter code generation model trained on 2T tokens (87% code, 13% language), offering state-of-the-art performance for multiple programming languages.
BRIEF DETAILS: A Helsinki-NLP Ukrainian-to-English translation model based on the Marian framework, achieving 64.1 BLEU score on Tatoeba test set.
BRIEF DETAILS: SigLIP base model (203M params) for vision-language tasks. Features sigmoid loss for better image-text understanding. Trained on WebLI dataset at 256x256 resolution.
BRIEF DETAILS: An advanced 7B parameter audio-language model capable of voice chat and audio analysis, supporting both speech interactions and audio signal processing with text instructions.
Brief Details: A sophisticated text-to-image model built on SDXL, featuring enhanced prompt understanding and creative capabilities with specialized fine-tuning using 220k GPTV captioned images.
Brief Details: VLM2Vec-Full is a 4.15B parameter vision-language embedding model based on Phi-3.5-vision-instruct, trained for unified multimodal embedding tasks.
Brief-details: Meta's 8B parameter LLM with strong performance on benchmarks - features 8k context length, BF16 precision, and outperforms many previous open source chat models
Brief Details: MeloTTS-Chinese is a high-quality text-to-speech model supporting mixed Chinese-English synthesis, optimized for CPU real-time inference with MIT license.
Brief-details: A Turkish-to-English translation model trained on OPUS data, achieving BLEU scores of 24.7-27.6 on news datasets and 63.5 on Tatoeba, using transformer architecture.
Brief-details: Legal domain-specific BERT model (small version) trained on 12GB of legal texts, offering 33% size of BERT-BASE with comparable performance and 4x faster inference.
BRIEF DETAILS: GPT-2 based text-to-image prompt generator fine-tuned on 250k Midjourney prompts, optimized for creative image generation instructions
Brief-details: Open-source 3B parameter LLaMA reproduction, trained on 1T tokens with RedPajama and Falcon datasets. Apache 2.0 licensed, suitable for text generation tasks.
